cpignata
commented
9 months ago Open cpignata opened 10 months ago
cpignata
commented
9 months ago 
ali-rezaki
commented
9 months ago Hi Jari,
Thanks a lot for your review of the Sustainability Considerations I-D. Your feedback has been very helpful. Much appreciated.
We are in the process of processing all your comments. Below, I wanted to share some responses to a few specific items.
Thanks & regards,
Ali
From: Jari Arkko [jari.arkko@gmail.com](mailto:jari.arkko@gmail.com) Sent: Thursday, February 15, 2024 2:35 PM To: e-impact@ietf.org; Carlos Pignataro [cpignata@gmail.com](mailto:cpignata@gmail.com); Ali Rezaki (Nokia) [ali.rezaki@nokia.com](mailto:ali.rezaki@nokia.com) Subject: Review of draft-cparsk-eimpact-sustainability-considerations Hi,
I have reviewed https://datatracker.ietf.org/doc/html/draft-cparsk-eimpact-sustainability-considerations-07 and had some thoughts.
Comments
First: thanks for this, it is a well written and useful document. [Ali: Thanks very much for your review and helpful comments.]
My other overall comment is that the document seems useful to publish, perhaps independent submission or through the program as an IAB stream document. But before that it would be useful to think about the scoping a bit. The current scope is somewhat large, and maybe a tighter scope would make the document easier to finish, give reader focus, etc.
"'this document also gives network, protocol, and application designers and implementors sustainability-related advice and guideance”
Maybe it would be better to focus network/protocol angles here. While there are some things in the document that can be classified as implementation-related, those parts are not particularly in-depth. You can basically state that ability to use various levels of sleep modes, have interfaces to getting data and managing power circuitry, and have interfaces that know about the type of power being applied. But I’m not sure you can do much more in this document, without significantly expanding the discussion and needing perhaps different type of expertise than we might have.
This document describes some of the tradeoffs that could be involved while optimizing for sustainability in addition to or in lieu of traditional metrics such as performance or availability. It also proposes some common terminology for discussing environmental impacts of Internet technologies, and gives network and protocol designers and implementors sustainability-related advice and guideance. Further, it discusses how Internet technologies can be used to help other fields become more sustainable.
There’s a lot there in general, even if the implementation aspects would be taken away. Perhaps a different packaging of the document would discuss help-other-fields and address-our-own-impacts separately. Perhaps making room for more dedicated discussion. Thoughts? [Ali: Thanks! Based on your comments and also what we have heard during the E-impact interim meeting, we are considering the different options of partitioning the document and focusing and sharpening each one. ]
- Definition of Terms
Great, much needed and well done.
Appropriate technology
I wondered to what extent this term is needed in this document. Does the definition come from somewhere? In any case, ”affordability” seems to be a clear issue not mentioned in the otherwise good list of considerations in the definition.
It is also not used in the rest of the document. [Ali: Appropriate technology, as a term and movement was started by the economist Fritz Schumacher as intermediate technology, in the 1970s. Earlier similar concepts also existed. Here is the Wikipedia article about it: Appropriate technology - Wikipedia . Affordability is indeed one of the aspects of this concept. One of the key messages here is that much of technology focus today is on high-end use cases and contexts, whereas, a global technology should not forget applicability to more basic/fundamental usages and much constrained economic and developmental contexts. A global technology for local contexts is indeed a target. In the I-D, while we strived to be inclusive of most of the relevant terms, we didn’t provide an example or usage for all the terms. Yet, one of the contemporary examples of Appropriate Technology development/use is provided by the “ACT4D: Appropriate Computing Technologies for Development” team of Prof. Aaditeshwar Seth in the Indian Institute of Technology in Delhi: ACT4D: Appropriate Computing Technologies for Development (iitd.ac.in) . We will add the references.]
As carbon intensity is location and time dependent, carbon awareness requires dynamic monitoring and response, such as carbon aware routing and networking
Perhaps in the definition we could avoid claiming that ca routing and networking is needed. For instance, I could argue that things that a major ca benefit could be acquired through ca-aware selection of compute, which indirectly also affects routing. Or maybe ”require” is too strong. [Ali: sure, we will change “require” to “could use”. ]
common outcome-oriented metric is energy consumption per data volume or traffic unit, in Wh/B [Telefonica]; this particular metric has however also been criticized for being easy to misinterpret, falsely indicating that systems are energy proportional even when they are not (see "Energy proportionality".)
Yeah. One issue is that communications tasks may differ from each other. Sending a thing from here to the other side of the planet, or to a non-near moving entity not tethered to anything are probably harder tasks than sending a packet on a local Ethernet or Bluetooth connection. [Ali: indeed, thanks. Here, we wanted to highlight the fact that we should avoid the misconception that reduced data volumes automatically lead to reduced network energy use and reduced GHG emissions. This proportionality is not currently observed in general.]
Handprint (environmental/ecological):
Really good to have this! [Ali: Thanks!]
Another fundamental sustainability impact assessment requirement is to cover the complete impact of a product, service or process over its full lifetime.
Is this related to internetworking? [Ali: Life Cycle Assessment is relevant to all parts of the ICT sector, including internetworking. Granted, most of the GHG impact of interworking is in its use phase but there are also other critical questions like when to replace an older product with a more energy efficient one (Total Cost of Ownership – TCO analyses) which can only be answered correctly by a LCA approach.]
3.2. Internetworking for Sustainability
Perhaps this too could be compressed a bit (see nits for other examples). But there’s another question mark. Perhaps SmartGrid, SmartCity, etc. are just examples of broader efforts, but there’s a lot also that doesn’t fit these categories, e.g., is XR/VR something that helps with sustainability or not? Answers might differ :-) Certainly virtual meeting tools would be … my point is that we perhaps don’t want to only list these broad, marketing-labels but also many other improvements can help with sustainability. Real-time information distribution about car traffic may reduce fuel or energy consumption, etc.
[Ali: There are indeed two key terms here, one is the use case, which determines the usage context of the technology, where smart grid, smart city are indeed very broad ones. The second is technology enablers that support the solution in the use case. Here, XR/VR or virtual meeting tools could be considered as technology enablers. While the handprint impact is being considered in a use case context, we need to include the footprint of the technology enablers in the equation to come to a balance of impact for the introduction of the technology solution in the use case process.]
Later addition: I see that S4 covers some of this. Maybe a bit fluffy there too though. [Ali: Thanks. We will work on the fluffiness in general 😊]
6.2. Multi-Objective Optimization
The fiber on/off links case is a simple and good example. But it also doesn’t highlight some of the inherent issues and design differences in general. For instance, it assumes that ”link off” implies link cannot be used for redundancy or for capacity. But this is highly dependent on how quickly such a link can be brought up and down. On relatively long up/down times, one wants to optimize for a stable condition and this raises also some of the tradeoffs. In a scenario where the times are much smaller, the equation is more of a dynamic consideration where you want to optimize how many packets a sudden traffic surge would cause to be queued and for how long, while you bring the link up again. In some cellular systems the timescales are small, so the question is more about whether the implementations can take advantage of this to save power between periods of packets sent. Secondly, there’s an issue of local vs. centralized control. If we look at the problem from the point of view of single interface card it may not be possible to do much, but with more central control (e.g., such as one might have in a base station among its antennas and transmitters, when the network is in charge of all scheduling decisions of when can different parties send) it may be possible to decide which resources are on and which are off, with more room for control than having to accept any packet that arrives at a given interface.
Perhaps some of this could be reflected here.
6.3.1. Redundancy and Sustainability 6.4. How Much are Performance and Quality of Experience Compromised?
One additional complicating factor is that there’s no single requirement for redundancy, performance, etc. — different classes of users or different applications may have different requirements, even some contractual ones.
One of the biggest potential opportunities in reducing environmental impact of networks concerns the ability to power resources such as equipment or line cards down when they are momentarily not needed due to swings in traffic demands. In general, this involves fully automated management control loops with very short time scales.’
Clearly .network management has a huge role. Including in the above. But it is not necessarily the only one driving some of this change. Local changes in the nodes through better implementation techniques can also take advantage of shutting down parts of equipment, typically on a much faster timescales than network management would be able to do. There’s of course a role for network management on longer time scales.
6.7.1. Metrics for Sustainability
A sustainability quantification framework is paramount for understanding the sustainability posture of a system, as well as its potential for aid in sustainability outcomes.
Would be nice to get more details of this, e.g., what metrics, is there infrastructure involved or just individual nodes producing information, etc. Or considerations relating to what information can or cannot be trusted.
(What’s the relation of this to material in Section 7 and onwards?)
Visibility: In this phase we focus on the measurement and collection of metrics. Insights and Recommendations: In this phase we focus on deriving insights and providing recommendations that can be acted upon manually over large time scales. Self-Optimization via Automation: In this phase we build awareness into the systems to automatically recognize opportunities for improvement and implement them.
Great summary.
Similarly, we also need to work on standard telemetry for collecting these metrics so that interoperability can be achieved in multi-vendor networks.
Yes. But maybe also intra- and inter-domain considerations need to be taken into account (as they may require different solutions). First phase is of course intra, but ...
Sustainability metrics and data models ought to describe how to secure the sustainability data exposed and surfaced through telemetry.
Falsification of advertised information from outside your own domain should be explicitly called out here as an issue as well.
Nits
s/guideance/guidance/
The impact of Internet technologies has been overwhelmingly positive over the past years (e.g., providing alternatives to travel, enabling remote and hybrid work, enabling technology-based endangered species conservation, etc.), and there is still room for improvement.
Should the and be a but? Or maybe ”…, but comes it with its own environment cost. There is a lot of room additional positive effects and reduced negative effects." [Ali: Agreed. We need to manage the negative impact of our technology while trying to maximize its usefulness and handprint. Will update.]
Another fundamental sustainability impact assessment requirement is to cover the complete impact of a product, service or process over its full lifetime. ..
This paragraph seemed a bit lengthy / duplicating a bit what was stated already earlier about LCA: [Ali: Agreed. Will optimize the text, removing repetitions.]
3.1. Sustainable Internetworking
Could perhaps be condensed since a lot of the material is already in the definitions. [Ali: Agreed. Overlpas with definitions will be replaced by references to definitions.]
6.6. Attributional and Consequential Models
Perhaps this could have come earlier in the document, when introducing terms and context. [Ali: actually, there is a reference from the LCA definition earlier in the document to this section but will check the flow.]
- Sustainability Guidelines for Protocol and Network Designers and Implementers
I think this section would generally read better if you focused on ”what do you need to document about sustainability aspects when designing a new protocol”, rather than focusing on a separate section. You can say there should be a section, or that a section is preferable, but focus first on what information needs to be thought through & documented.
Jari
Process all these useful e-impact mailing-list reviews